ADRepository

Dataset Overview

Numerical Datasets

• Source: From the KDD19 paper – DevNet
• Count: 7 datasets
• Basic statistics:
  • Dataset: donors, census, fraud, celeba, backdoor, campaign, thyroid
  • Data size: 619,326, 299,285, 284,807, 202,599, 95,329, 41,188, 7,200
  • Dimensionality: 10, 500, 29, 39, 196, 62, 21
• Details & benchmarks: Available in the DevNet paper; code at the DevNet GitHub repository.

Categorical Datasets

• Count: 14 datasets
• Basic statistics:
  • Dataset: bank, census, AID362, w7a, CMC, APAS, CelebA, Chess, AD, Solar‑flare, Probe, U2R, R10, CoverType
  • Data size: 41,188, 299,285, 4,279, 49,749, 1,473, 12,695, 202,599, 28,056, 3,279, 1,066, 64,759, 60,821, 12,897, 581,012
  • Dimensionality: 10, 33, 114, 300, 8, 64, 39, 6, 1,555, 11, 6, 6, 100, 44
  • Anomaly class: yes, 50K+, active, yes, child>10, train, bald, zero, ad., F, attack, attack, corn, cottonwood
• Details & benchmarks: Provided in the corresponding papers.

Video Datasets

• Datasets: ShanghaiTech Campus, UCF‑Crime
• Characteristics: Features extracted with I3D backbone, suitable for weakly‑supervised video anomaly detection.
• Details: Available in the related publications.

Image Datasets

• Count: 14 datasets
• Application domains: Defect detection, novelty detection, lesion detection in medical images, anomaly segmentation in autonomous driving scenarios.
• Details: Available in the related publications.

Graph Datasets

• Graph‑level anomaly detection: 16 datasets (e.g., PROTEINS_full, ENZYMES, AIDS, …).
• Node‑level anomaly detection: 4 datasets (e.g., YelpRes, YelpHotel, YelpNYC, Amazon).
• Details: Provided in the related literature.

Time‑Series Datasets

• Common datasets: ASD, SMD, SWAT, WaQ, DSADS, Epilepsy
• Details: Refer to the associated papers.
• Note: Existing datasets have certain usage issues; consult the cited works for best practices.